Automatic groove identification in 3D bullet land scans
Kiegan Rice
Automatic groove identification
in 3D bullet land scans
Kiegan Rice
Iowa State University
February 21st, 2018
Background: Bullet Lands
When a gun is fired, the bullet is propelled forward through the barrel
As it travels down the barrel, it makes contact with parts of the barrel
Striations result from this contact
Striations are observed on
lands
. Lands are separated by
grooves
.
Background: Collection of Land Data
Sensofar Confocal Light Microscope
Hamby set 44 - 35 bullets from 10 consecutively rifled Ruger barrels
Rescanned with CSAFE’s microscope
To use as ‘base’ set; methods to be tested on several other data sets
Have manual groove identifications for this whole set
Each pixel: .645 square microns
Each land is 2mm (2000 microns) wide
A scan of one bullet (6 individual lands) takes ~1 hour
Background: Land Surface
Importance of Groove Removal
Bullet matching algorithm
Removing the underlying curved structure of land
Looking at remaining residuals
Deviations from the natural curve of the land
Importance of Groove Removal
Error Rates
Traditional rifling and manual identification of grooves
Bullet-to-bullet matching error rate is 0.
Polygonal rifling and coated bullets
Error rate is higher
Biggest error source: misidentification of groove locations
Shortcomings of Current Methods
Robust Linear Models
Groove Identification Process
Once we have the residuals, how do we decide where the cutoffs should be?
Comparing two automated approaches
We will look at all the residuals in the areas between our predicted grooves and the manually identified grooves
For each land in the data set (208 total), sum up these residuals
Comparing two automated approaches
The sums shown previously are calculated for each land and each method
Then, we compare the distributions of those values
Impact for Forensic Analysts
Currently: Individual case studies
Collaborations with different police depts. and forensic institutes
Phoenix PD, Los Angeles PD, St. Louis PD, Denver PD
Story County Sheriff’s Office (IA)
Houston Forensic Science Center
Impact for Forensic Analysts
Future goals:
Get algorithm certified
no certification process in place for admissibility
Adhere to OSAC’s updated standards for 3D measurements (Feb. 9, 2018)
Support firearms examiners in their job
Address concerns raised by NRC 2009 and PCAST 2016
removing subjectivity from the assessment
allow for quantification of error rates
Acknowledgments
All work was
sponsored
by CSAFE (Center for Statistics and Applications in Forensic Evidence), a NIST Center of Excellence
Work
advised
by Drs. Heike Hofmann and Ulrike Genschel of CSAFE/Iowa State University